126: Chapter 126 Low Entropy Workshop, Open for Business

July 10th, 9:27 AM.

Without any warning, Low Entropy Workshop updated with a new video.

When the video was posted, 99% of the initial viewers were not from the robotics community, automation majors, or geek circles.

Rather, they were ordinary netizens and mathematics enthusiasts drawn by the trending name, Jiang Lin.

Because the Low Entropy Workshop account had already been thoroughly dug into countless times.

Everyone knew that Low Entropy Workshop was Jiang Lin.

They clicked in almost with the mindset of catching his response.

Full of expectation, some had even prepared screenshot tools, expecting to see some ground-shaking mathematical declaration or a blackboard covered with continued fractions and probability measure formulas.

Or at least, they expected to see Jiang Lin say in a tone cool to the point of being sharp, "Regarding the PFR, I will provide formal text later."

Instead, when the video loaded, there was no face of Jiang Lin on screen, no high-definition screenshots of the ICM Report, and not even a single mathematical symbol.

There were only pure white words on a black background, bearing the cold, hard feel of an industrial instrument.

[G-01: Non-Periodic Embodied Mobile Platform Concept Demonstration]

In the next second, the scene cut in.

What appeared was not Jiang Lin's face, nor mathematical formulas, let alone a screenshot of the ICM Report.

Instead, it was a foot end made of white polyoxymethylene resin.

It had no cosmetic rendering whatsoever, revealing its bare mechanical structure.

High-strength aluminum alloy connecting rods, exposed wiring harnesses, and reducer casings with a metallic sheen at the joints.

The robotic foot slowly descended from the top of the frame, pressing onto the surface of an obstacle board sloped at approximately 15 degrees.

On the right side of the screen, a complex real-time data panel was displayed synchronously.

Terminal Control Log.

Contact curve plot begins.

Z-axis force feedback: 0N → 12N → 45N → 88N.

The elastic polyurethane layer at the bottom of the foot underwent millimeter-scale compressive deformation.

At the moment of contact, a slight deflection occurred in the normal direction of the foot to conform to the inclined plane.

The force control loop intervened, and the contact curve stabilized within milliseconds without any overshoot spikes.

In the bottom left corner, green subtitles popped up silently.

[Foot Contact Confirmed: Confidence 99.7%]

[Local Stability Zone Calculated]

[Stance Phase Updated]

At that moment, an eerie two-second vacuum fell over the bullet comments.

Then, it erupted like a tsunami.

[?????]

[Who am I? Where am I? Did I click the wrong video?]

[Wasn't he supposed to solve the PFR Conjecture? Why did he post a robot? That's quite a leap!]

[Wait, is this the thing rumors said was haunting—ah no, that Jiang Lin was tinkering with at night at Jiangda University's Engineering Training Center?]

[Is Low Entropy Workshop really Jiang Lin himself? Isn't he a pure mathematician?]

[Can an expert explain what this plain white leg is doing? Testing pressure?]

...

But the video showed no intention of catering to the audience or stopping to explain.

In the second clip, the camera angle raised, switching to an overhead view.

On the floor lay twelve obstacle boards with entirely different materials, shapes, and inclination angles.

This was not a smooth runway in a standard lab, nor was it a meticulously designed, highly regular flight of stairs like Boston Dynamics uses.

It was a maliciously designed unstructured test track.

The height difference of each obstacle board fluctuated randomly between 5 to 15 centimeters.

The tilt angles ranged from -20 degrees to +25 degrees.

The spacing was completely irregular—some were right next to each other, while others required a leap.

Furthermore, the surface friction coefficients were entirely different.

Some were unpolished, rough pine boards.

Some were cold-rolled metal plates covered with a layer of fine sand.

Some were acrylic plates with slight protrusions along the edges, resembling damaged steps.

There was even one board with a light dusting of talcum powder on its surface.

The subtitles displayed the test conditions in an emotionless, standardized engineering format.

[Non-Repetitive Obstacle Board Group A]

[Terrain Topology Parameters: Globally Unknown]

[Gait Controller: No Preset Periodic Template]

[Test Objective: Low-Speed Autonomous Blind Navigation]

Then, the screen split into two down the middle.

Left side: [Conventional Periodic Gait Baseline Control]

Right side: [G-01 Non-Periodic Stance Phase Dynamic Control]

The audience finally saw the full picture.

The hexapod platform on the left was a baseline unit loaded with a conventional control algorithm based on the same physical mechanical structure.

Its movements looked very orderly, following classic tripod or wave gaits.

Using fixed phase offsets, its footholds advanced according to an internal preset metronome like soldiers marching in step.

The baseline machine smoothly passed the first two relatively gentle obstacle boards.

But when it reached the third board, which had an 18-degree incline, trouble arose.

The internal clock of the baseline machine called for it to place its front-right foot down at this moment.

However, due to the cumulative height error from the first two boards, the foot dropped not in the center of the wooden board, but straight into mid-air over the edge.

The foot sensor instantly reported no ground contact.

Following conventional algorithms, the controller attempted to forcibly extend the leg downward to seek support, but the body had already incurred an irreversible side tilt as its center of gravity shifted forward.

The Zero-Moment Point within the support polygon rapidly slipped out of the safety boundary.

In the video, the body lurched violently, the joint motors emitted a harsh screech of overload, and then the entire machine lost power, collapsing limply onto the obstacle board.

A glaring line of red text popped up on the screen.

[Periodic Gait: Center of Mass Instability, Protective Shutdown Triggered]

The video did not skip the failure.

On the second attempt, the baseline machine switched control strategies, adjusting its stride length and forward speed.

This time it survived the third board, but at the fifth board—the sand-covered metal plate—a minor physical sideslip occurred the moment the middle-right foot made contact.

For a periodic gait with fixed timing, a minor displacement of a single foot meant the phase of the entire dynamic model was disrupted.

High-frequency oscillation shook the body, and to prevent the reducer gears from shattering, the system cut off the power once again.

[Periodic Gait: High-Frequency Phase Trajectory Divergence, Protective Shutdown Triggered]

Third attempt: stepping frequency drastically reduced.

The result: on the step board with edge protrusions, the leg lift height was insufficient; the front foot caught on the protrusion, triggering a forward-roll dynamic singularity and breaking down for the third time.

Three tests, all interrupted.

Faced with continuous, unknown, and unstructured composite terrain, the conventional algorithms exhibited fragile robustness.

The left side of the screen froze on the remains of the third failure and dimmed.

The right side of the screen lit up.

G-01, this pure mechanical entity stripped of all cosmetic covers, began to power up.

The first leg—the front-right foot—slowly descended.

Not according to a preset time, but relying on high-frequency end-effector force feedback.

The moment the POM foot touched the first wooden board and the force value reached the threshold, the joints smoothly switched from position control to impedance control.

Contact confirmed.

The second leg—the middle-left foot—followed.

G-01's gait looked almost weird.

It lacked any sense of rhythm.

The stance phase did not switch according to a fixed timing at all; instead, only after each foot confirmed solid placement and calculated the maximum friction cone provided by the current support surface did it solve in real time which leg should move next.

It smoothly crossed the first two boards.

Arriving at the third board—the inclined board where the baseline machine first fell—G-01 raised its front-right foot, preparing to plant it down.

Yet less than two centimeters from the board's edge, it paused slightly.

The entire body hung suspended in mid-air for roughly 0.4 seconds.

To ordinary viewers, this might look like a video lag or the machine freezing up.

But for the terminal log scrolling synchronously on the right, an invisible storm of computing power occurred within those 0.4 seconds.

The logs flooded the screen frantically.

[Foothold Prediction: Edge Slip Risk 87%]

[Current Swing Phase Interrupted]

[Local Stability Zone Recalculated]

[Local Contact Map Updated]

[Stance Phase Reschedule Triggered]

G-01 did not step down as originally planned.

The hovering front-right foot retracted three centimeters backward, gently settling at the base of the inclined board.

At the same time, the rear-left leg, which was supposed to be a supporting leg, did not continue to bear weight. Instead, breaking the fixed rhythm, it reached forward early to step firmly on the flat ground to the side, forcefully adding a new support point for the body at the critical moment and expanding the support margin.

The body dipped slightly due to this sudden shift in center of gravity, but immediately after, the internal Model Predictive Control algorithm kicked in instantly, with each joint motor outputting precisely calculated compensation torque to level the body's posture once again.

No fall.

No high-frequency oscillation either.

G-01 smoothly crossed the third obstacle board.

The footage that followed was like a completely one-sided dimensionality strike.

Fifth board: the sand-covered metal plate.

The moment the foot detected a sudden drop in friction coefficient upon contact, G-01 did not force power through it; instead, it instantly reduced tangential thrust on that leg and redistributed power to the three legs standing on stable ground.

Passed.

Step board with edge protrusion: the front foot touched the obstacle, generating a tiny collision reaction force.

The algorithm instantly judged it an unexpected contact, re-planned the foot trajectory in real time, raised it by five centimeters, and stepped over the protrusion.

Passed.

Final board: a mixed-height, extremely treacherous staggered board.

Passed as well.

On the right side of the screen, there were only lines of green results.

[Non-Repetitive Obstacle Board Group A: PASSED]

[Low-Speed Stance Phase Dynamic Switching: VERIFIED]

[Lateral Micro-Perturbation Recovery: VERIFIED]

[Abnormal Slip Protection Link: ONLINE]

[Time Elapsed: 41.2s | Status: SYSTEM STABLE]

This time, even ordinary viewers who understood nothing about dynamic equations or Jacobian matrices fully got the point.

On the left: following rules, trying to fight chaos with fixed rhythms, resulting in three falls.

On the right: abandoning rules, playing it safe step by step amidst chaos, calculating while moving, and passing smoothly in a single try.

After a brief moment of processing, the comment section exploded completely.

[Holy crap! Holy crap! Three shutdowns on the left, while the right just waded right through!]

[I get it! The left is a blind man marching in step according to a fixed routine, while the right is a minesweeping sapper who has his eyes closed but probes with every single step!]

[It's not moving slowly; it's frantically calculating! That 0.4-second pause was godly—it predicted it would fall and swapped legs on the fly to brace itself!]

[What kind of black technology is this? Abandoning periodic gait? Isn't that a carbon-based creature's instinct? Isn't that exactly how we walk on a pile of loose rocks?]

[So non-periodicity wasn't just a theoretical gimmick—he actually built it?]

[If this thing were scaled up and placed in earthquake ruins, or on the surface of the Moon or Mars, can you even imagine how huge of a deal that would be?]

The video continued.

The third section was a hardcore close-up of the terminal logs, showing the jumping code and state machines.

[IMU Status: Normal]

[Foot Contact Force Feedback: 1000Hz Closed-Loop, Normal]

[Pawl Physical Locking Mechanism: Normal]

[Joint Phase Current: No Abnormal Spikes]

[Drive Bus Voltage: Stable]

[Reducer Backlash Compensation: Normal]

The fourth section was a lateral physical perturbation test that made many know-it-alls hold their breath.

G-01 stood stably across two boards of uneven height.

Suddenly, a push rod equipped with a pneumatic device extended from off-screen, striking the side of the obstacle board without warning.

The chassis shook violently.

The contact stress at G-01's front-right foot underwent an avalanche-like shift in an instant, losing all load-bearing capacity.

In baseline periodic mode, under an unexpected lateral impact of this magnitude, the system's state space would be breached instantly, with the only outcome being motor protective lockup and a complete machine roll-over.

But G-01 did not collapse immediately, nor did it attempt to forcibly resist the impact force like traditional robots, which usually leads to even greater reverse oscillations.

Yielding to the impact force, its body tilted sideways by about 15 degrees.

Time seemed to slow down.

It paused briefly.

The joints throughout its body seemed to shed rigidity in an instant, entering a low-impedance compliance mode, maintaining only the minimum support torque required to stay within posture boundaries.

[part:gemini-3.5-flash-lite]

Within a millisecond, based on the angular velocity and acceleration data from the IMU, the internal controller recalculated the heavily deformed local stability zone under the current tilt state.

Then, a miraculous operation occurred.

Instead of trying to stand up, the left rear support leg fell heavily outward in advance, forming a new pivot point.

At the same time, the load-less right foot retracted instantly, actively releasing all torque output to avoid becoming a pivot point for leverage effects.

The moment the new support topology was established, the motors throughout the body redistributed torque.

On the verge of falling to the limit, the machine's posture was forcefully pulled back into the stable range by a precisely calculated resultant force.

Three hundred milliseconds.

From being hit, losing stability, reconstructing support, to recovering balance, it took only three hundred milliseconds.

After three hundred milliseconds, the motors hummed, and as if nothing had happened, it took another step forward.

The subtitle was only one line, yet it was filled with a sense of academic brutality that looked down upon the world.

[ Aperiodic Support Phase Redistribution: Lyapunov Margin Certificate Passed Online ]

In the fifth segment, the video finally provided a panoramic view.

G-01 slowly crawled onto the final, highest, and most rugged obstacle board, stopping steadily on the finish line.

Its six legs presented an irregular yet absolutely stable support posture.

The last page was still that unshakeable white text on a black background.

[ G-01 Public Demo v0.1 ]

[ Aperiodic Embodied Mobile Platform ]

[ Low Entropy Workshop ]

Total video duration, one minute and fifty-nine seconds.

There were also only three short engineering notes in the description.

[ Phase 1: Foot Contact Link and Actuator Verification. ]

[ Phase 2: Aperiodic Support Phase and Foot Contact State Evaluation. ]

[ Note: This video is only a public concept prototype demonstration, without dustproof, waterproof, or explosion-proof design, and is not a final engineering version. ]

...

In fact, after Jiang Lin clicked upload, he directly closed the browser page.

He focused his attention on another screen, where the GPU cluster was running mathematical models at full capacity.

Of course, he was also very clear that this one-minute and fifty-nine-second video would definitely not be quiet.

Because the global robotics circle in July 2022 was inherently a battlefield full of anxiety and involution, never lacking in excitement.

Over the past few years, Boston Dynamics had stood like a towering mountain before all robotics practitioners.

With their yellow quadruped robot dog Spot and bipedal humanoid robot Atlas, they cruelly refreshed the public's imagination of the upper limits of robotic movement capabilities.

Spot had already entered preliminary commercialization scenarios in fields such as factory inspection and hazardous material handling.

Atlas even performed anti-human extreme actions such as parkour, backflips, jumping, and single-leg balance in official videos.

They had almost become the sole industry benchmark in the public mind for how well a robot could move.

All startups would be asked by investors during roadshows: ‐How do you compare to Boston Dynamics?‐

Meanwhile, domestically, the intensive release of quadruped robots like Unitree's Go1 series and Xiaomi's CyberDog, relying on terrifying cost-control capabilities, pushed multi-legged robots from lofty laboratories, expensive military orders, and hardcore geek circles into the consumer-grade market.

Speed, following, obstacle avoidance, companionship, open-source developer platforms...

These commercially enticing keywords were repeatedly plastered by various manufacturers onto colorful promotional pages and dazzling short video titles.

Further away in Silicon Valley, Tesla's humanoid robot Optimus hung over the entire industry like a massive and oppressive boot.

In July 2022, everyone was holding their breath, waiting to see what physical objects Musk would bring out at the upcoming AI Day.

The capital market had already begun to revel, with countless research reports arguing.

If Tesla could seamlessly migrate its visual perception system, FSD computing power system, and terrifying gigafactory manufacturing system accumulated in the field of autonomous driving to humanoid robots, would the script of the next industrial revolution be directly rewritten?

Therefore, when Low Entropy Workshop, a completely unfamiliar account, pushed the video of G-01 to the homepage of the robotics section on major platforms, the first reaction of many industry insiders was not shock, but a certain instinctive exhaustion and contempt.

‐Another walking machine.‐

‐Another quadruped or hexapod platform that just modifies an open-source solution to come out for financing.‐

‐Another college student's graduation project trying to prove that I can also make a robot by putting together a few pieces of wood.‐

Until they patiently watched the entire one minute and fifty-nine seconds.

In the eyes of senior industry experts, the nature of the matter began to undergo a drastic chemical reaction.

G-01 did not demonstrate any high-speed running capability.

It did not demonstrate any anthropomorphic or pet-like interactive actions, such as shaking hands or bowing.

It did not put on any approachable, cute shell to lower the audience's vigilance.

It didn't even have footage of running like a dog behind its owner in a park, which was the easiest to spread and attract traffic across the entire web.

It just slowly, clumsily, and even stutteringly completed an extremely foundational task on a set of grey-looking non-repetitive obstacle boards.

Without pre-setting any periodic gait templates, relying purely on a low-level tactile closed-loop, it continuously confirmed foot contact, dynamically redistributed support phases, and steadily delivered the machine body to the destination when subjected to external non-linear impacts.

Experts know the ropes.

This was not comparing engineering maturity with Spot at all.

It was not comparing the hydraulic-driven extreme dynamic capabilities with Atlas.

It was not comparing cost reduction, efficiency increase, and consumer-grade market penetration with Unitree and Xiaomi.

Still less was it comparing those illusory general humanoid capital stories with Tesla.

What it tapped into was a path that very few people in the industry dared to touch, the easiest for ordinary audiences to ignore, yet the only compulsory path toward true embodied intelligence.

The underlying local survivability in unstructured, complex, and unknown terrains.

11:10 AM.

The first review with true academic weight did not come from a network influencer, but from a heavyweight scholar who had been deeply engaged in the foundation of domestic legged-locomotion robots for years.

Professor Xu Mingchuan.

The head of the legged robot team at a national key laboratory of a top C9 university, and a Yangtze River Scholar.

He was unquestionably the domestic authority in the fields of multi-joint dynamic modeling and non-linear control.

He directly dropped the video link of one minute and fifty-nine seconds into the research group's core chat group, which had always been on mute.

The group contained over a dozen doctoral students and postdocs.

Unusually, instead of sending long-winded guidance, he only typed out two short yet weighty sentences.

[ This is not those hexapod toys running on flat ground using open-source MPC algorithms. ]

[ He completely bypassed the gait library of traditional dynamics. ]

A few minutes later, Chen Yang, a postdoc who had been struggling with Boston Dynamics' early papers, couldn't help asking.

[ Teacher Lu, what does bypassing the gait library mean? Does it mean offline trajectory planning is no longer done? Then how is its center-of-mass momentum kept stable? ]

Professor Xu Mingchuan replied exceptionally quickly, obviously staring at the screen all along.

[ Look closely at the 0.4-second pause at the 41st second. Most demonstration videos on the market either showcase the extreme torque of the actuators, such as how high they can jump, or the stability of the controller under a certain set of preset known terrain tasks. Even when Spot climbs stairs, what the public sees is a set of highly engineered and maturely packaged task capabilities. ]

[ But what G-01 showcases is that under extreme conditions where the normal vector, friction force, and elevation of each contact surface are all completely random and non-repeatable, its support phase can actually break away from a fixed time template and achieve spatial re-closure. ]

About ten seconds later, Professor Xu Mingchuan added a comment that made all doctoral students' hearts leap with fright.

[ If the terminal logs of this video are truly synchronized and not edited or falsified in post-production, this technical route is enough to make our laboratory reassess half of the priorities we wagered on fixed-phase optimization over the past three years. Save the video immediately, hold a group meeting at 1 PM this afternoon, and deconstruct its phase space trajectory frame by frame. ]

After this sentence was screenshotted and leaked from the internal group of the national key laboratory, it quickly sparked a storm in automation institutes and robotics groups across major universities.

It carried far more weight than a million ordinary netizens dropping expletives.

Because Professor Xu Mingchuan was not an attention-seeking traffic blogger, nor was he a gawking mechanical enthusiast.

He knew very well that for a legged platform pieced together from aluminum alloy and motors to slowly walk across a few wooden boards, there was nothing technically novel about it in itself; plenty of student works could achieve that.

What truly terrified him was that this hexapod aperiodic mobile platform actually walked based on a brand-new underlying control logic akin to biological instinct.

The second node to suffer a fierce impact came from the industry.

At an ongoing monthly R&D summary meeting of a top-three domestic special robotics company, the head of the motion control department directly interrupted a subordinate who was reporting on the progress of SLAM algorithms, and cast the video of G-01 directly onto the massive 85-inch monitor in the conference room.

Sixteen people sat in the conference room, including the hardware director, software director, and product vice president.

The video played once.

The product vice president frowned, his tone somewhat puzzled: ‐Is this made by that Jiang Lin who's been popular online recently? Isn't the speed too slow? How can this turtle-like crawling efficiency be used in industrial inspection? Customers value coverage area. And this shell has nothing done to it, with wire ends exposed outside.‐

The head of motion control stood in front of the screen and did not immediately refute.

Holding a laser pointer, he precisely rewound the progress bar to the segment of the lateral physical perturbation test.

Pause.

Zoom in on the dense terminal logs on the right.

Press the play button.

Bang!

The wooden board was struck, the machine body tilted sideways, the legs reorganized, and balance was restored.

The head of motion control turned around, his gaze sweeping across everyone at the long table.

‐Exterior design is worthless; we have the best industrial designers who can pump out ten sets of shells in a week. Absolute speed is also worthless; replacing them with higher-power brushless motors and optimizing the reduction ratio can increase the speed.‐

‐At this stage, picking on it for lacking a commercial product shell and maturity is an extremely foolish perspective.‐

He circled the 0.4-second pause log in the picture with the crimson dot of his laser pointer.

[ Local Stability Zone Recalculation ].

‐What's truly valuable is right here.‐

‐Gentlemen, for our current quadruped products, if we encounter this level of unexpected lateral slippage, besides cutting off the power and letting the dog lie down, is there a second strategy?‐

The conference room instantly fell into silence.

The software director pushed up his glasses and shook his head with a solemn expression.

‐G-01 did it. Completely devoid of a periodic gait template as a time anchor, it managed to maintain the dynamic closed-loop of the support phase purely through a burst of computing power in an extremely short time.‐

The head of motion control took a deep breath.

‐This is equivalent to the machine reinventing a standing posture mid-air while falling.‐

The product vice president finally realized the seriousness of the matter, leaning forward slightly: ‐What do you mean?‐

‐If the content of this video is not CG, not faked,‐ the person in charge put down the laser pointer and said decisively, ‐then the person who wrote this control framework surpasses our entire team present here in their capability regarding low-level non-linear optimization algorithms.‐

‐So, contact him immediately?‐ asked the person beside him.

‐Contact, contact immediately.‐

‐Under what name, should the investment department go talk?‐

‐Don't use those arrogant investment tones, and don't talk about any acquisitions. The other party is Jiang Lin, the Jiang Lin who presented at that ICM. Talking to him about acquisition right off the bat is an insult to him.‐

The head of motion control immediately stopped him.

"First, we'll go under the guise of technical exchange, and our attitude must be sincere. We need to figure out how its foot-end contact model is constructed, and how it solves the real-time computing power bottleneck of nonlinear optimization."

13:00 in the afternoon.

Three and a half hours after the video was released.

The business contact email for Low Entropy Workshop was already stuffed to the brim.

The first formal official email came precisely from the university laboratory where Professor Xu Mingchuan worked.

The subject line was extremely rigorous: [Invitation to Technical Exchange on Aperiodic Gait Algorithms and Unstructured Contact Modeling]

The wording of the body text was polite and respectful.

[Respected Low Entropy Workshop: We have carefully studied the demonstration video of G-01. We hope to conduct a closed-door technical seminar with you regarding foot-end contact modeling under non-repetitive terrain, dynamic reallocation of support phases, and underlying safety state machine design.]

At the end of the email, a sentence full of sincerity was specially attached.

[Considering that the technology you demonstrated may be at the forefront of the industry, if it involves undisclosed core details or code, we are very willing to sign the highest-level two-way non-disclosure agreement before the exchange.]

If the first email represented academia's thirst for underlying theory, the emails that followed began to imbue matters with a raw, gritty sense of reality.

The second email came from a well-known domestic mining energy inspection equipment company.

They skipped the pleasantries and technical flattery, sending three shocking on-site exploration photos directly in the attachment.

The first photo: A dark, pitch-black mine scree slope located hundreds of meters underground. The slope was over thirty degrees, and the ground was covered with sharp rock fragments of varying sizes and distinct edges.

The second photo: An abandoned, damp roadway. The ground was full of muddy water, several broken railway ties haphazardly blocked the middle of the road, and rusted wire mesh hung over the ties.

The third photo: A narrow equipment angle passage. The ground was mixedly covered with concrete blocks, scattered rebar, and fine coal cinder.

The body of the email was brief and filled with a helpless industrial appeal.

[Hello Mr. Jiang, as you can see, this is the real operating environment we currently face.]

[When our existing high-configuration wheeled inspection platform enters the scree slope in the first picture, the tires are extremely prone to sinking and slipping, making it completely impossible to pass stably.]

[We tried switching to a heavy-duty tracked platform. Although the pass rate increased somewhat, in the complex terrains of the second and third pictures, the track gaps easily get jammed with coal cinder and rusted wire, causing the machinery to lock up, resulting in extremely high maintenance costs.]

[As for the mature quadruped robot solutions currently on the market, not only is the procurement cost high, but when dealing with such unpredictable, random, soft contact surfaces, they frequently trigger protective shutdowns, making it impossible to form a continuous inspection work stream.]

[If the content of the G-01 video released by your esteemed party is true, and its aperiodic support reallocation capability truly possesses the characteristics of resisting unstructured terrain, we very much hope that you can arrange a closed-site test targeting real terrain samples. All testing and prototype modification expenses will be borne by our side.]

Jiang Lin sat in front of the computer, and the mouse cursor lingered over those three photos for a long time.

This was not the entertaining "holy shit" from Bilibili bullet comments.

It was not exaggerated praise written by tech media trying to hitch a ride on the trend.

It was not tentative communication from academic peers.

This was the Real World—a Real World full of random destructiveness that didn't care about any theoretical formulas—handing an on-site test paper directly to G-01.

Gravel.

Deep stagnant water.

Metal obstacles and dust.

Narrow, enclosed spaces with dynamic constraints.

One hundred percent non-repeatable contact.

This environment, of course, still fell far short of the ruins Jiang Lin had experienced in the Wasteland World deep within his mind.

But at least it was no longer the playhouse-style test environment that Jiang Lin had manually laid out using a few bought wooden boards on the flat cement floor of the Engineering Training Center.

It was the true current state of the industry, and it would also be the first step for G-01 to walk out of the greenhouse.

Besides these two emails, there was a massive volume of other information in the inbox.

The third email came from Sequoia Capital, a top domestic venture capital institution.

The subject line was filled with capital's arrogance and directness.

[Emergency communication regarding the possibility of early angel round financing for the Low Entropy Workshop robot project — Looking forward to a talk with Mr. Jiang]

Jiang Lin only glanced at the subject line, didn't even click open the email, and directly right-clicked to mark it as low priority / temporarily unhandled.

The fourth came from several mainstream tech media outlets, including 36Kr, Huxiu, and others.

They hoped to conduct an exclusive in-depth interview on the explosive topic of a mathematical genius crossing over to build a revolutionary robot.

Temporarily unhandled.

The fifth one made Jiang Lin ponder for a brief moment.

That was a private email from a partner at a top domestic patent agency and intellectual property legal institution.

The other party's acuity was very high, and his words went straight to the crux of the matter.

[Mr. Jiang Lin, apologies for the intrusion. As a patent lawyer who has long focused on hardware innovation, I must remind you. The public video you released already constitutes a certain degree of public disclosure in a legal sense. Although the video does not show the internal code, some technical features can already be reverse-engineered from the external motion characteristics. If your esteemed party subsequently wishes to carry out global PCT international applications, Paris Convention priority rights, and claims layout for core control methods in core fields such as aperiodic control architectures, foot-end contact reallocation algorithms, and underlying safety state machines, I suggest immediately stopping the release of any videos containing underlying UI and terminal logs, and conducting a confidentiality evaluation on patent search and application strategies with us as soon as possible. Otherwise, it is extremely easy for overseas competitors to preemptively register peripheral patent pools using minor modifications.]

This email put an inescapable problem right on the table.

Low Entropy Workshop could not remain just a video account forever.

Before releasing the video, Low Entropy Workshop was merely an outlet for Jiang Lin to verify his technology.

But after this video detonated the entire industry, it had to immediately transform into a sturdy vessel capable of resisting reality's shocks and carrying massive commercial and legal rules.

13:16 in the afternoon.

The cell phone placed next to the desktop keyboard began to vibrate.

Caller ID: Lu Zhixing.

Jiang Lin pressed the answer button.

"I've seen the video."

Professor Lu Zhixing's voice carried an expected exhaustion and an irrepressible excitement.

"This time it's not just your small circle and the comment section that are lively; the matter has broken out of the circle. Just now, in an expert group of a think tank under the Ministry of Industry and Information Technology, someone forwarded the video in. Professor Xu Mingchuan has seen it too, did you know?"

"I just found out." Jiang Lin tapped the keyboard, dragging the file into the database.

"He's a person with immense sway in this direction domestically when it comes to legged-foot robots. If he thinks favorably of it, many national-level project resources will open up to you."

Jiang Lin opened the browser, quickly searched for Professor Xu Mingchuan's publication records, and his gaze swept over the titles of several papers on nonlinear MPC and whole-body control.

Hearing the rhythmic mechanical keyboard sounds coming from the other end of the phone, Professor Lu Zhixing blurted out, "You're not searching his profile right now, are you?"

"Yes, his research direction leans more towards traditional model-based optimization, and his adaptive capability in unpredictable environments is insufficient." Jiang Lin objectively evaluated.

Professor Lu Zhixing was choked up by Jiang Lin's tone, which bordered on academic judgment, and was momentarily at a loss for words.

After a good two seconds, he said helplessly, "His evaluation of you is extremely high. Not only the academic circle, but enterprises over there will also immediately come looking for you like sharks smelling blood."

"They've already come," Jiang Lin said.

"University labs?"

"They've come."

"Investment term sheets from venture capital firms?"

"A few have come."

"Interview requests from major media outlets?"

"All sorts, probably."

Professor Lu Zhixing took a deep breath, his tone turning serious: "Jiang Lin, you are currently in the eye of the storm. The one-minute-and-fifty-nine-second video has pushed you to a crossroads that is extremely dangerous yet full of opportunities. How do you plan to handle this table full of hot properties?"

Jiang Lin did not answer immediately.

On the screen, he was ruthlessly classifying the emails.

[University Exchange] — Classified as academic communication.

[Enterprise Testing (Mining)] — Marked in red, highest priority.

[Patents and Legal] — Marked in yellow, urgently needs processing.

[Investment Institutions] — Cold palace.

[Media PR] — Deleted.

Subsequently, he skillfully opened an Excel spreadsheet named: [Low-Entropy Business Log].

Column 1: Source institution.

Column 2: Demand type (technology / business / interview).

Column 3: Whether real harsh scenario datasets are provided.

Column 4: Whether willing to sign a two-way NDA, test data ownership agreement, and non-exclusive verification agreement.

Column 5: Whether the exchange involves underlying core control frameworks (resolutely blocked).

Column 6: Whether a legal entity is required to sign.

Column 7: Priority.

Jiang Lin unhesitatingly marked the email from the mining inspection robot company as P0 — the highest priority.

It was not because this company might give the most money, nor because the commercial value was the greatest.

But because it provided the dirtiest, most chaotic real terrain that was hardest to directly fit with mathematical formulas.

In front of that scree slope, any fancy algorithms and praises appeared pale and powerless.

Feedback data from real terrain was more precious than the citation counts of ten top-journal papers combined.

He said to Professor Lu Zhixing on the other end of the phone, "Teacher Lu, I need to register a corporate entity."

Professor Lu Zhixing did not feel surprised on the other end of the phone; he even let out a long sigh of relief.

"You guy, who is immersed in the mathematical world, have finally begun to acknowledge the operating rules of Real World society."

Jiang Lin did not pick up on this evaluation, instead putting forward his own thoughts in a steady speaking pace.

"I indeed need a shell, a sturdy entity capable of doing the following three things."

"First, to be able to sign the most stringent non-disclosure agreements and on-site test waiver agreements on equal footing."

"Second, to be able to legally hold all future patent pools, control method patents, software-hardware collaborative structure patents, software copyrights for code implementation, and undisclosed trade secrets to build a moat."

"Third, to be able to set up a corporate public account to undertake subsequent high-spec hardware procurement, outsourced precision machining, and the inevitable small-scale hiring of personnel in the future."

Professor Lu Zhixing listened carefully and hit the nail on the head, pointing out the core issue: "You want a corporate entity, no problem. But maintaining hardware R&D and on-site testing at this level burns through money like crazy. How many of those venture capitals that came knocking do you plan to see, and what range do you intend to set for the financing valuation?"

"Not for now." Jiang Lin's answer lacked the slightest hesitation.

"Why? Aren't you short of money right now?" Professor Lu Zhixing was puzzled.

"I am, but not to the point where I need to sell off the company's control at this stage," Jiang Lin said.

It wasn't that he didn't need money, but that he didn't need equity financing.

Quantitative accounts, outsourcing settlements, and technical service contracts were enough to tide G-01 over its most dangerous early verification period.

"That statement is quite sober-minded."

Professor Lu Zhixing sighed on the other end of the phone.

He had heard of many tragedies where talented entrepreneurs compromised with capital due to depleted funds and ultimately lost control of their projects.